解析雪花中的json键值，键中有一个点

Question

{
  "deviceLocale": "en_US",
  "deviceSerialNumber": "xxxxxxxxxx",
  "eventSource": "abc",
  "ext.user.browser": "Mobile Safari",
  "ext.user.browser.version": "1.0.4",
  "ext.user.device.family": "iPhone", 
  "ext.user.os": "iOS",
  "ext.user.os.version": "1.3.0",
  "Timestamp": 158007896874 }

This is the sample json that I have..

Parsing in snowflake

```
select distinct
eve_id,
json_payload:ext.useragent.device.family::varchar as type,
json_payload:ext.useragent.os::varchar as osname,
json_payload:ext.useragent.os.version::varchar as os 
from XYZ table, lateral flatten (input => json_payload)
```

But all these three fields are giving NULL values and I see the data in the json format. So I guess parsing is not right. I know in snowflake while parsing if we use a dot or : then it refers to the nested key. But in my case I have a simple json with no nested keys.

Any idea?

Answer 1

firstly you can put the name in double quotes, like so:

SELECT parse_json('{
        "deviceLocale": "en_US",
        "deviceSerialNumber": "xxxxxxxxxx",
        "eventSource": "abc",
        "ext.user.browser": "Mobile Safari",
        "ext.user.browser.version": "1.0.4",
        "ext.user.device.family": "iPhone", 
        "ext.user.os": "iOS",
        "ext.user.os.version": "1.3.0",
        "Timestamp": 158007896874 }') AS json_payload,
    json_payload:"ext.user.device.family"::varchar as type,
    json_payload:"ext.user.os"::varchar as osname,
    json_payload:"ext.user.os.version"::varchar as os;

gives:

JSON_PAYLOAD    TYPE    OSNAME  OS
{    "Timestamp": 158007896874,    "deviceLocale": "en_US",    "deviceSerialNumber": "xxxxxxxxxx",    "eventSource": "abc",    "ext.user.browser": "Mobile Safari",    "ext.user.browser.version": "1.0.4",    "ext.user.device.family": "iPhone",    "ext.user.os": "iOS",    "ext.user.os.version": "1.3.0"  }    iPhone  iOS 1.3.0

OR you can use the [''] format like json_payload['ext.user.os.version']::varchar as os which allow the avoidance of double quotes (in case you want to avoid that).

In you access SQL you have json_payload:ext.useragent.device.family::varchar but the useragent section is only user in your JSON. So that will give you trouble.

Also in your example you are using a LATERAL FLATTEN, but asking how to access the flattened members of the object you are flattening. So the flatten is not needed. But if you are want to flatting then you will get a row per top level item, at which point you would be want to filter the key .. but I suspect this is not what your trying to do. But if you are is a good idea to alias the flatten to help show the intention.

WITH jp AS (
    SELECT parse_json('{
      "deviceLocale": "en_US",
      "deviceSerialNumber": "xxxxxxxxxx",
      "eventSource": "abc",
      "ext.user.browser": "Mobile Safari",
      "ext.user.browser.version": "1.0.4",
      "ext.user.device.family": "iPhone", 
      "ext.user.os": "iOS",
      "ext.user.os.version": "1.3.0",
      "Timestamp": 158007896874 }') AS json_payload
)
SELECT 
    f.key,
    f.path,
    f.value
FROM jp, LATERAL FLATTEN (input => json_payload) f;

gives:

KEY PATH    VALUE
Timestamp   Timestamp   158007896874
deviceLocale    deviceLocale    "en_US"
deviceSerialNumber  deviceSerialNumber  "xxxxxxxxxx"
eventSource eventSource "abc"
ext.user.browser    ['ext.user.browser']    "Mobile Safari"
ext.user.browser.version    ['ext.user.browser.version']    "1.0.4"
ext.user.device.family  ['ext.user.device.family']  "iPhone"
ext.user.os ['ext.user.os'] "iOS"
ext.user.os.version ['ext.user.os.version'] "1.3.0"